OpenAIRE
Refine
Year of publication
Document Type
- Article (168)
- Doctoral Thesis (19)
- Review (18)
- Preprint (16)
- Bachelor Thesis (10)
- Other digital publishing (9)
- Book (8)
- Master's Thesis (6)
- Bookcollection (4)
- Part of Bookcollection (2)
Keywords
- - (129)
- Computer Science, general (18)
- IT in Business (16)
- Information and Computing Sciences (6)
- Künstliche Intelligenz (6)
- machine learning (5)
- Nachhaltigkeit (4)
- Scattering (4)
- Soziale Arbeit (4)
- Studium (4)
Institute
Data Governance
(2026)
Laminar jet apparatuses (LJAs) are employed to study gas–liquid mass transfer and thereby provide insights into the design of technical separation and reaction equipment. The hydrodynamics of the laminar jet determines the operating window of such apparatuses and must therefore be understood for the proper evaluation of experimental results. Although substantial literature exists on free liquid jets, quantitative analyses directly tailored to LJAs and their operating conditions remain scarce. The present work addresses this gap. We investigate all hydrodynamic aspects relevant to the operation of LJAs using computational fluid dynamics (CFD) simulations: the shape of the jet surface (i.e. the gas–liquid interface), the velocity profile within the jet (including the transition from the nozzle to the free jet) and jet breakup. The study is conducted for a square-edged orifice plate nozzle under various flow conditions and analysed using dimensionless numbers. The CFD simulations account for gravity and surface tension and comprise the entire relevant domain (feed line, nozzle, liquid jet and surrounding gas phase). Complementary experiments using a LJA, in which the jet shape was measured, validate the simulation results. The findings provide quantitative guidance for rational design of LJAs and offer a foundation for replacing ad hoc assumptions in the evaluation of the experiments.
A commonly employed imaging modality that relies on ionizing radiation is Computed Tomography (CT). While lowering the radiation dose is beneficial for patient health, it can result in reduced image quality. Therefore, improving low-dose CT (LDCT) reconstruction is a significant area of research. The LoDoPaB-CT benchmark evaluates LDCT reconstruction methods, where many top methods use UNet-type architectures. We explore a two-stage approach for LDCT reconstruction: the first stage employs traditional filtered backprojection (FBP), while the second stage performs CT image enhancement. Our training strategy involves pretraining a neural network to denoise natural grayscale images, which are corrupted by Gaussian noise, followed by fine-tuning the network for CT image enhancement using LDCT and normal-dose CT (NDCT) pairs. Experiments on various small subsets of the LoDoPaB-CT dataset demonstrate the effectiveness of our method, showing that less task-specific data are required for training.
In Magnetic Particle Imaging (MPI), the scans of delta concentrations can be collected in a system matrix and a target distribution of particles injected in a specimen can be retrieved by regularized inversion of the associated linear system, using the scan of the specimen as data. Recent publications show that ill-posed inverse problems can be solved with Plug-and-Play (PnP) algorithms, which split at each iteration the general regularized inversion into a simpler Tikhonov-type problem and a Gaussian denoising problem. By substituting the Gaussian denoising step with machine learning-based denoisers, it is possible to leverage the performance of Neural Networks in the denoising task. In particular, it has been shown that it is possible to employ publicly available and general-purpose denoisers into the MPI reconstruction task in a Zero-Shot fashion (no ad hoc training necessary). In the specific algorithm considered, each denoising step takes as input the noise level of the Tikhonov subproblem and works in particular with a very coarse estimation of the noise level as variance of the iterate. In this work we explore the benefit of using a better estimation of the noise level using convolution with the Laplacian.
Urban sound has long been managed primarily within a noise-control framework. To support a more resource-oriented planning perspective, reproducible city-scale screening tools are needed to identify where spatial-physical conditions may support soundscape-relevant service-supply pathways and inform early-stage planning prioritisation. This study introduces the Urban Soundscape Service-supply Potential (USSP) framework as a proxy-based, open-data and planning-oriented approach for mapping such conditions without treating them as direct measures of actual acoustic conditions, perceived soundscape experience or realised cultural ecosystem service benefits. Using Munich, Germany, as a testbed, four proxy-based sound-source potential layers were derived from open geospatial data: traffic and industrial sound-source carrier potentials, human-activity sound-source opportunity potential and natural sound-source support potential. These layers were integrated into three screening-level indices representing naturalness-oriented (USSPN), liveliness-oriented (USSPL) and balance-oriented service-supply potential (USSPB). The results revealed distinct spatial configurations of pressure-related and resource-related potentials, while district-level trade-off typologies clarified relative noise-resource conditions for planning diagnosis. At the composite level, USSPN showed a marked core-periphery contrast, USSPL exhibited a heterogeneous mosaic, and USSPB showed a stronger concentration in upper equal-interval classes. All three indices displayed significant positive spatial autocorrelation and local hotspot and coldspot structures. The eBird-based ecological plausibility check provided limited external support for USSPN by showing a positive association with bird species richness after controlling for sampling effort. USSP complements, rather than replaces, perceptual soundscape assessment and exposure-based noise governance by supporting early-stage prioritisation, targeted field verification and planning discussion.
Low-dose and sparse-angle computed tomography (CT) reduces radiation exposure but makes image reconstruction challenging due to noisy and limited projection data. Popular reconstruction methods are based on two-stage approaches, typically involving filtered backprojection (FBP) followed by a neural network to enhance the image. FBP, however, amplifies noise and struggles with irregular sampling. Therefore, we explore filter-free initial reconstructions, shifting the filtering step to the neural network. In particular, we investigate how two-stage methods can be adapted for cases where implementing explicit filters is difficult, such as with irregular sampling. Specifically, we propose backprojection (BP) or a small number of Landweber iterations as the initial reconstruction, followed by a fine-tuned DRUNet model, referred to as BP-DRUNet and Landweber-DRUNet, respectively. For evaluation, we consider both regular and irregular sampling conditions: For regular sampling, we compare BP-DRUNet with FBP-DRUNet (using FBP as the initial stage) in order to benchmark against standard two-stage approaches. BP-DRUNet performs comparably to FBP-DRUNet under regular sampling. In irregular sampling, Landweber-DRUNet improves reconstruction quality with more iterations, though at the cost of longer training and inference times. Experiments are carried out on synthetic and real CT datasets with parallel- and fan-beam acquisitions across different sparse-angle setups.
It is well known that the performance of the Benjamini-Hochberg (BH) procedure can be improved by incorporating estimators of the number or proportion of null hypotheses to yield an adaptive BH procedure which still controls FDR. Several such plug-in estimators have been proposed. For some of these, such as Storey’s estimator, plug-in FDR control has been established, while for some others, such as the Pounds-Cheng estimator, some gaps remain to be closed. These developments have largely focused on the case of continuous test statistics, where null p -values follow the uniform distribution. In the discrete setting, although these estimators continue to provide plug-in FDR control, they become overly conservative, leading to inefficient procedures. In this paper, a general class of estimators that encompasses the classical Storey and Pounds-Cheng estimators is introduced. Alongside, several generic strategies to mitigate conservativeness in the discrete setting are proposed by incorporating information about the null distribution functions. These strategies provably yield less conservative estimates while maintaining valid FDR control, and the resulting performance gains are illustrated on both real and simulated data. As a byproduct of a more general result, plug-in FDR control for the Pounds-Cheng estimator in the continuous case is also established.
Because of the observation that kinetic rates can be better correlated with in vivo drug efficacy than simple affinities, significant efforts have been made to develop experimental and computational methods to predict drug–target residence times over the past few years. Here, we summarize the discussions and reflections from an international Centre Européen de Calcul Atomique et Moléculaire (CECAM) Workshop, which took place in March 2025, with participants from both academia and industry, including experimentalists as well as computational scientists, and was dedicated to computational methods for protein–ligand binding kinetics prediction. As currently standing challenges, we identify the need for standardized benchmark datasets, the need to move from simple to more complex and biomedically relevant targets and the question of how academia and industry can work together to move the field of drug–target binding kinetics forward. We also discuss what level of accuracy can be expected from computational methods and highlight that the field would benefit from blind challenges to enable a fair comparison of different computational methods to predict kinetic rates.
We experimentally demonstrate that pairs of time-delayed ultrabright and ultrashort X-ray pulses of two different colors, delivered by modern X-ray Free Electron Lasers, can provide two time-delayed snapshots of a sample. We introduce Dichography, a method that algorithmically separates the diffraction signals overlapping on the detector and independently retrieves the two images of the specimen. We employ Dichography to reconstruct two views of individual xenon-doped helium nanodroplets with 20 nm spatial resolution. The consistency of structures observed in both images at delays up to 750 fs provides evidence that, under these illumination conditions, significant structural damage only occurs at longer timescales. We further validate the method by imaging pairs of silver nanoparticles intercepted by the same light pulse. Dichography enables a new class of experiments across physics, chemistry, and materials science, making a significant step toward the original promise of X-ray free-electron lasers to capture ultrafast movies of nanomatter.
